General_Effort@lemmy.world 6 months ago
When I post something here, I usually have the hope that someone else benefits from it in some small, tiny way. Either because it makes them laugh or because it gives them valuable information.
I don’t understand why anyone would post if they can’t bear the idea that someone else might get a tiny, little something for free.
Power nor wealth doesn’t come from a single individual, but from a great many.
Just because you vote, say or do something, doesn’t mean it’ll change much of anything. But if hundreds, thousands, millions, or billions do, then we have change. It’s exactly the same with data. A single data-point is not worth very much.
If we all work together, we can make sure that none of us can benefit from the other? How does that even make sense?
Look. I am unable to understand why this bothers you. I like feeling I have a positive effect on the world. I like knowing, EG, that my taxes help the less fortunate. What you are saying seems completely absurd to me.
Your comment assumes everyone agrees ChatGPT or other LMMs/etc are a net benefit to society.
I’m on the fence about it myself.
There’s no such assumption. The people who use it must feel it benefits them, or they wouldn’t use it. By the same logic, the makers must be getting something out of it, as well.
If there is some net negative for society that I am not seeing, then I don’t see why I should even be allowed to offer my work for AI training. Normally, harming other people is a big “NO”.
FaceDeer@kbin.social 6 months ago
It really bothers me how normalized the "screw everyone else, I want to get paid!" Attitude has become. OP is just assuming that everyone agrees with it. I want AI to be well trained. If self-interest must be assumed then consider how useful AI is for you.
And copyright doesn't even apply here in any event. Training an AI is not copying, it is learning.
Osa-Eris-Xero512@kbin.social 6 months ago
I think a lot of the concern here, for me if noone else, is them taking the data and then turning it around into a closed for-sale product. If AI is going to be trained, it should be trained well, but if the result of doing so is them turning around and charging [me/us/everyone, as applicable] an ass load for the privilege of its use then I want no part of it.
AI trained on public data should be public. So if adding boilerplate is the solution to this problem, let it be infectious licensing which forces opening of the resultant model to the public.
General_Effort@lemmy.world 6 months ago
It’s not the solution. It looks more like the problem to me.
As has been said, licensing doesn’t work. When you write something, you automatically have the copyright. With a license, you allow others to copy it, which otherwise would be illegal. For this to work, copyright would have to be extended to cover learning. That’s obviously terrible. But let’s assume for a moment that copyright is only extended to cover machine learning (ML).
You would not be allowed to use anything on the internet for ML, unless there was a license allowing it. It would basically outlaw scraping the net for training data. You’d have to sift through everything to find stuff, with a permissive license. Of course, no for-profit enterprise would pick anything up with an infectious license.
AI companies would have to pay for training licenses. Non-profits that cannot pay would be limited to public domain data: Stuff that is so very old that it is out-of-copyright, some government publications and, of course, your infectiously licensed posts. Sound good?
The for-profit stuff would be more expensive to generate a steady cash-flow to “rights-holders”, like streaming does today.
Well, maybe that’s what you want. For some people, this is simply a matter of ideology. They feel that (intellectual) property is supposed to work like that, and damn the consequences. I’m going to assume you are not like that.
We’d create a new, steady flow of money to property owners, who have to do nothing in return. It’s nice to be rich. I don’t think we need to make it nicer, but that would be the result.
It would be great for corporations like Microsoft that already have a lot of intellectual property for training. It would also be great for the likes of Meta, that can just amend their TOS to get a license from their users. Traditional publishers would likely also see a nice windfall profit, as they’d be able to sell all their old newspapers, magazines and books.
To me, this just seems crazy. It’s doubling down on everything that’s already going wrong.
I’m guessing that that is not the outcome you want. So, the question would be, how you came to support a policy that would lead to it.
FaceDeer@kbin.social 6 months ago
Even private for-pay AI is useful to me. Even ones I don't pay for myself, since other AI developers have been making heavy use of existing AI models to generate data for training their own new models.
In any event, as I said, copyright doesn't even apply here. Adding a CC license does nothing, it's not "infectious" to AI models trained off of it.
Osa-Eris-Xero512@kbin.social 6 months ago
Yeah, CC doesn't cover it in any case. Any attempt would probably need some sort of bespoke license to specifically target the training use case while still allowing comments to be used like normal.
And a Microsoft-sized pile of money to fight it out in court.